aboutsummaryrefslogtreecommitdiffstatshomepage
path: root/cgi-bin/get.py
diff options
context:
space:
mode:
authorEgor Tensin <Egor.Tensin@gmail.com>2021-03-03 23:09:11 +0300
committerEgor Tensin <Egor.Tensin@gmail.com>2021-03-03 23:09:11 +0300
commit69580ef09cfb0f41fa50d3252f4a83cfb7f153b5 (patch)
tree87f2c4c07719f30698a0f34c61020c395181c7d3 /cgi-bin/get.py
parentprune obsolete entry from .gitignore (diff)
downloadlinux-status-69580ef09cfb0f41fa50d3252f4a83cfb7f153b5.tar.gz
linux-status-69580ef09cfb0f41fa50d3252f4a83cfb7f153b5.zip
add server.py
It runs a web server and imports the request handling classes from get.py directly, hence eliminating the performance culprit (which was the `import` processing for each request).
Diffstat (limited to 'cgi-bin/get.py')
-rwxr-xr-xcgi-bin/get.py46
1 files changed, 40 insertions, 6 deletions
diff --git a/cgi-bin/get.py b/cgi-bin/get.py
index 3831812..1453dc2 100755
--- a/cgi-bin/get.py
+++ b/cgi-bin/get.py
@@ -10,6 +10,7 @@ import cgi
from collections import namedtuple
from concurrent.futures import ThreadPoolExecutor
from enum import Enum
+import http.server
import json
import os
import pwd
@@ -38,16 +39,43 @@ class Response:
def __init__(self, data):
self.data = data
- def print(self):
- print("Content-Type: text/html; charset=utf-8")
- print()
- if self.data is not None:
- print(self.dump_json(self.data))
+ def headers(self):
+ yield 'Content-Type', 'text/html; charset=utf-8'
@staticmethod
def dump_json(data):
return json.dumps(data, ensure_ascii=False)
+ def body(self):
+ return self.dump_json(self.data)
+
+ def write_as_cgi_script(self):
+ self.write_headers_as_cgi_script()
+ self.write_body_as_cgi_script()
+
+ def write_headers_as_cgi_script(self):
+ for name, val in self.headers():
+ print(f'{name}: {val}')
+ print()
+
+ def write_body_as_cgi_script(self):
+ if self.data is not None:
+ print(self.body())
+
+ def write_as_request_handler(self, handler):
+ handler.send_response(http.server.HTTPStatus.OK)
+ self.write_headers_as_request_handler(handler)
+ self.write_body_as_request_handler(handler)
+
+ def write_headers_as_request_handler(self, handler):
+ for name, val in self.headers():
+ handler.send_header(name, val)
+ handler.end_headers()
+
+ def write_body_as_request_handler(self, handler):
+ if self.data is not None:
+ handler.wfile.write(self.body().encode(errors='replace'))
+
def run_do(*args, **kwargs):
output = subprocess.run(args, stdin=DEVNULL, stdout=PIPE, stderr=STDOUT, universal_newlines=True, **kwargs)
@@ -318,6 +346,12 @@ class Request(Enum):
def __str__(self):
return self.value
+ @staticmethod
+ def from_http_path(path):
+ if not path or path[0] != '/':
+ raise ValueError('HTTP path must start with a forward slash /')
+ return Request(path[1:])
+
def process(self):
if self is Request.STATUS:
return StatusTask().complete()
@@ -331,7 +365,7 @@ class Request(Enum):
def process_cgi_request():
params = cgi.FieldStorage()
what = params['what'].value
- Request(what).process().print()
+ Request(what).process().write_as_cgi_script()
def main():